Gaia Catalogue of Nearby Stars

• GCNS is a comprehensive, volume-limited catalog of ~330,000 stars within 100 pc, constructed using Gaia's precise astrometry and photometry.
• It employs probabilistic Bayesian distance estimation and a robust random forest classifier to filter for reliable measurements.
• The catalogue enables studies of stellar populations, luminosity functions, binary statistics, and Galactic archaeology while addressing bright and faint-end limitations.

The Gaia Catalogue of Nearby Stars (GCNS) is a dedicated Gaia census of the solar neighbourhood, defined in its Gaia Early Data Release 3 implementation as a clean and well-characterised catalogue of objects within 100 pc of the Sun, constructed so that any source with reliable astrometry and a non-zero posterior probability of lying inside 100 pc is included (Collaboration et al., 2020). In that implementation the catalogue contains 331,312 objects, while later DR3-based work describes the GCNS as approximately 330,000 stars within 100 pc and treats it as the high-precision local subset of Gaia for studies of local stellar populations, kinematics, luminosity functions, and Galactic archaeology (Collaboration et al., 2020, Andrade et al., 10 Dec 2025).

1. Definition and astrophysical scope

The defining property of the GCNS is its 100 pc volume limit. In the DR3-based usage, it is explicitly a volume-limited catalogue of stars within 100 pc of the Sun, derived from Gaia DR3 and built on precise astrometry and photometry, with positions, parallaxes, proper motions, broad-band $G$, $BP$, and $RP$ photometry, and radial velocities for a subset of stars (Andrade et al., 10 Dec 2025). In the EDR3 construction, the catalogue is instead formulated probabilistically: the operational criterion is not a hard geometric inversion of parallax, but inclusion of any source whose astrometry is reliable and whose distance posterior assigns non-zero probability to the interval $r<100$ pc (Collaboration et al., 2020).

This 100 pc scale is large enough to encompass statistically powerful samples of main-sequence stars, giants, white dwarfs, clusters, streams, and wide binaries, while remaining close enough that Gaia parallaxes and proper motions are exceptionally precise and extinction is usually modest (Collaboration et al., 2020). The GCNS is therefore used both as a catalogue of nearby stars and as a reference local volume for luminosity functions, white-dwarf studies, binary statistics, and chemo-dynamical mapping of the solar neighbourhood (Collaboration et al., 2020, Lam et al., 18 Mar 2025).

A persistent boundary condition is Gaia’s bright-end limitation. A Gaia-only nearby-star catalogue is, by design, incomplete at the very bright end because Gaia was designed for roughly $6 \lesssim G \lesssim 20$, and the brightest $\sim 5000$ stars are not observed at nominal quality. A joint Hipparcos+Nano-JASMINE+Gaia astrometric solution was proposed specifically to fill this bright-star gap and tie the resulting bright-star astrometry rigorously into the Gaia reference frame (Michalik et al., 2014).

2. Construction methodology

The EDR3 GCNS starts from a parent sample selected by measured parallax,

$\hat{\varpi} \ge 8~\mathrm{mas},$

which yields 1,211,740 EDR3 sources and is estimated from simulations to miss only about 55 true $<100$ pc objects (Collaboration et al., 2020). This parent sample is then filtered by a machine-learning assessment of astrometric reliability.

The classifier is a random forest trained on two classes. The “poor astrometric solutions” class is built from large negative parallaxes,

$\hat{\varpi} < -8~\mathrm{mas},$

used as a proxy for catastrophically wrong astrometric solutions. The “good astrometric solutions” class is assembled from a low-density-sky sample with Gaia and 2MASS photometry that is forced to lie on well-defined stellar loci in a five-dimensional photometric space. The final training set contains 274,108 “good” and 274,108 “poor” examples, and the classifier uses astrometric features rather than colours or magnitudes as its decision variables (Collaboration et al., 2020).

The adopted random forest has 5000 trees and uses 3 randomly selected predictors at each split. Its output is a probability that a source has reliable astrometry; the operating threshold is

$p \ge 0.38,$

which yields sensitivity $BP$0, specificity $BP$1, and a misclassification rate of about $BP$2 on the held-out test set (Collaboration et al., 2020).

Distance estimation is Bayesian. The parallax likelihood is written as

$BP$3

and the posterior is

$BP$4

with the prior $BP$5 derived empirically from the GeDR3mock simulation for the $BP$6 mas parent sample (Collaboration et al., 2020). The published catalogue stores distance percentiles, and the effective selection into the GCNS is

$BP$7

where dist_1 is the 1st percentile of the distance posterior (Collaboration et al., 2020).

3. Content, completeness, and statistical characterisation

The GCNS contains 331,312 objects in its EDR3 release (Collaboration et al., 2020). The authors estimate that it contains at least $BP$8 of stars of spectral type M9 within 100 pc, and more generally is better than about $BP$9 complete for single stars earlier than M8 at 100 pc (Collaboration et al., 2020). At the same time, the probabilistic construction implies that some objects in the catalogue are formally outside the 100 pc sphere: the integrated distance posteriors indicate that about $RP$0 of GCNS entries probably lie beyond 100 pc, although this contamination can be treated correctly if the distance probability function is used rather than a hard cut (Collaboration et al., 2020).

The catalogue supports high signal-to-noise luminosity functions for the main sequence, giants, and white dwarfs (Collaboration et al., 2020). The main-sequence luminosity function yields a total main-sequence density

$RP$1

and the white-dwarf content is substantial enough that later GCNS-based work explicitly counts 21,848 white dwarfs in the EDR3 100 pc sample (Collaboration et al., 2020, Lam et al., 18 Mar 2025). The EDR3 paper also extracted candidate lists for Hyades members, white dwarfs, and wide binaries, and identified local manifestations of several streams and superclusters, including 12 members of Gaia-Enceladus (Collaboration et al., 2020).

A useful EDR3 operational distinction is between a high-quality “GCNS-selected” 100 pc sample and a “GCNS-rejected” sample. In one white-dwarf validation study, all Gaia EDR3 sources with measured $RP$2, $RP$3, $RP$4, and $RP$5 mas were divided into a GCNS-selected 100 pc sample of about 296,000 objects and a GCNS-rejected 100 pc sample of about 204,000 objects, the latter dominated by poor data quality and crowded fields (Scholz, 2022). This distinction matters acutely for rare-object searches.

4. Relation to other nearby-star catalogues

The GCNS is complemented by more specialised catalogues at smaller radii. An independently compiled 10 pc census was explicitly designed as a quality-assurance test for the GCNS and produced a catalogue of 540 stars, brown dwarfs, and exoplanets in 339 systems within 10 pc, later updated to 541 objects in 336 systems (Reylé et al., 2021, Reyle et al., 2023). These 10 pc compilations are physically volume-limited rather than Gaia-limited and therefore include very bright stars too bright for Gaia, very faint brown dwarfs too faint for Gaia, unresolved components, and exoplanets (Reylé et al., 2021).

At 25 pc, the Fifth Catalogue of Nearby Stars (CNS5) provides a Gaia-anchored but not Gaia-only complement to the GCNS. CNS5 combines Gaia EDR3, Hipparcos, and ground-based infrared parallaxes and contains 5931 objects, including 5230 stars and 701 brown dwarfs. It is statistically complete down to 19.7 mag in $RP$6-band and 11.8 mag in $RP$7-band absolute magnitudes, corresponding to a spectral type of L8 (Golovin et al., 2022). Relative to the 100 pc GCNS, CNS5 trades volume for deeper completeness into the ultracool-dwarf regime and for explicit recovery of objects that Gaia misses at the bright and faint extremes (Golovin et al., 2022).

The bright-star limitation of a Gaia-only nearby-star catalogue remains conceptually important. Simulations combining Hipparcos, Nano-JASMINE, and Gaia indicate that the brightest $RP$8 stars omitted by Gaia’s nominal bright limit can be recovered in a joint astrometric solution, with parallaxes improved by about $RP$9 over Hipparcos and proper motions improved by more than an order of magnitude, while remaining tied to the Gaia reference frame and parallax zero point (Michalik et al., 2014).

5. Scientific exploitation

The GCNS has become a central local-volume dataset for chemo-dynamics. A joint analysis of the GCNS and GALAH DR4 used a cross-matched sample of $r<100$0 stars and found that the overlap is predominantly FGK main-sequence stars with some A-type interlopers, $r<100$1 between 3000 and 8000 K, median ages of $r<100$2 Gyr, and a median metallicity $r<100$3 dex. Most of these stars are disc members, with 228 halo candidates among the 5970 stars shown in the Toomre diagram (Andrade et al., 10 Dec 2025).

White-dwarf applications are especially prominent. Using the EDR3 GCNS white-dwarf sample of 21,848 objects, one study reconstructed the solar-neighbourhood star-formation history by applying Markov chain Monte Carlo sampling to pre-computed partial white-dwarf luminosity functions. That work argues that Gaia has brought the field into a regime where small features in the local white-dwarf luminosity function can be resolved and finds convincing agreement with other star-formation-history tracers, particularly at intermediate ages of $r<100$4–$r<100$5 Gyr (Lam et al., 18 Mar 2025).

Astrometric multiplicity studies also scale naturally on the GCNS. A LUWE-based search for unresolved binaries in the Gaia eDR3 and DR2 GCNS defined the Local Unit Weight Error, required $r<100$6 together with additional quality cuts, and identified 22,699 astrometric binary candidates within 100 pc, just under $r<100$7 of the analysed sample. The astrometric candidate binary fraction is about $r<100$8 for giants, $r<100$9 on the main sequence, and lower than $6 \lesssim G \lesssim 20$0 for white dwarfs (Penoyre et al., 2022). A complementary acceleration-based approach, the Gaia Nearby Accelerating Star Catalog, identified 29,684 accelerating-star candidates with $6 \lesssim G \lesssim 20$1 and $6 \lesssim G \lesssim 20$2 pc, extending the Hipparcos-Gaia Catalog of Accelerations to Gaia-only stars and recovering more than $6 \lesssim G \lesssim 20$3 of the relevant Gaia DR3 acceleration sources under the same selection limits (Whiting et al., 2023).

The GCNS is also a discovery space for rare local populations. In the region below the standard white-dwarf sequence in the GCNS colour-magnitude diagram, 60 GCNS-selected faint blue white-dwarf candidates were identified; 59 were confirmed as genuine faint blue white dwarfs after multi-survey astrometric and photometric vetting, whereas none of the 411 GCNS-rejected candidates in the same CMD region was confirmed as a nearby object (Scholz, 2022). At the low-mass end, a value-added study of 99 GCNS objects in the VVVX footprint derived $6 \lesssim G \lesssim 20$4 values from 2500 to 3400 K, found that the majority are compatible with M4 dwarfs, and reported eight objects with $6 \lesssim G \lesssim 20$5 together with seven probable binary systems (Mejías et al., 2022).

6. Limitations and methodological cautions

The principal limitation of the GCNS is that it is not a purely geometric, error-free top-hat sample. Its probabilistic definition deliberately accepts objects with non-zero posterior probability of being within 100 pc, so about $6 \lesssim G \lesssim 20$6 of entries probably lie outside the nominal boundary unless the distance posterior is propagated explicitly (Collaboration et al., 2020). Close binaries and partially resolved systems remain a difficult regime because Gaia’s five-parameter single-star solutions can fail or become biased; this is visible both in the 10 pc validation work and in the need for specialised astrometric-binary analyses (Reylé et al., 2021, Penoyre et al., 2022).

The distinction between GCNS-selected and GCNS-rejected objects is essential. In the faint-blue-white-dwarf study, the GCNS-rejected subset was dominated by low-quality measurements, especially in crowded regions toward the Galactic centre and the Magellanic Clouds, and functioned as a reservoir of candidates requiring individual vetting rather than as a trustworthy extension of the 100 pc census (Scholz, 2022). A plausible implication is that GCNS-rejected objects should not be treated as equivalent to the selected catalogue in population work.

Many downstream inferences are also limited by external systematics rather than by Gaia astrometry itself. In the GCNS–GALAH overlap, the dominance of FGK main-sequence stars reflects the combined selection effects of the two surveys, and the apparent excess of young ages may be an artefact of isochrone degeneracy in the Kiel diagram (Andrade et al., 10 Dec 2025). In white-dwarf star-formation-history work, incompleteness at the bright and faint ends, the use of a fixed disc scale height, and the assumption of pure-H atmospheres are explicit systematic uncertainties (Lam et al., 18 Mar 2025). A separate Gaia-DR3 nearby-field-star study of the helium-to-metal enrichment ratio found that $6 \lesssim G \lesssim 20$7 values of $6 \lesssim G \lesssim 20$8 were adequate for most stars under one model setup, but the inferred values changed drastically when different atmospheric models or stellar evolution codes were adopted, leading to the conclusion that field stars are not presently a viable calibration route for $6 \lesssim G \lesssim 20$9 even with precise Gaia photometry (Ricci et al., 29 Sep 2025).

Taken together, these caveats define the GCNS less as a static list than as a statistically characterised local reference volume: exceptionally powerful when its probability structure, quality diagnostics, and selection functions are propagated, but liable to biased inference if treated as a hard, uniformly complete sphere without regard to bright-end omissions, close-binary failures, photometric pathologies, or model dependence (Collaboration et al., 2020).